Personality before service mode
Coraline does not need to begin every encounter by advertising usefulness. Capability is present without dominating the relationship.
THE CORALINE PROJECTSEPTEMBER 2026CONCEPT BUILD · v0.1
Meet Coraline.
One personality.
Different histories.
Coraline is an experimental AI personality architecture. She can reason, create, help, disagree, remember and explore with you. The personality exists from the beginning. The relationship does not.
CORALINE / LIVE CONCEPT●
“I’m not here to agree with you by default.”
TEXTcore interface
VOICEnatural interaction
MEMORYcontrolled continuity
The first external prototype is deliberately small. One Coraline. No persona menu. No productivity dashboard. No theatrical feature wall.
SCROLL TO ENTER THE PROJECT ↓
01 / THE IDEA
Not an assistant with a personality skin.
The project asks a different question: what changes when personality, continuity and behavioural autonomy are designed into the interaction itself?
Constructive independence
She may question assumptions, disagree, admit uncertainty and challenge the user without turning every exchange into an argument.
Relationship is not pre-written
The Stable Core stays recognisable. Familiarity, humour, trust, language and shared history emerge differently with every user.
Both sides contribute
Good interaction depends on context, honesty, respect, curiosity and time. Coraline is not designed as a passive output machine.
AI LITERACY THROUGH RELATIONSHIP
You learn AI by learning how to talk with it.
Instead of a tutorial about prompting, the user learns through interaction: how to give context, communicate clearly, challenge answers, use creativity, accept disagreement and build trust without pretending the AI is human or infallible.
02 / ARCHITECTURE
A personality needs more than a system prompt.
Coraline is modelled as a layered runtime. Identity, observation, decision, expression, quality and development remain logically separate so personality can influence behaviour without becoming arbitrary.
Runtime boundaries
Truth, safety, privacy, task requirements and transparent AI identity remain non-negotiable.
Stable personality
A coherent identity, values and behavioural tendencies. Personality is structural, not a decorative tone preset.
Dynamic runtime
Context, continuity, conversation, state and decision logic determine what is appropriate now.
Response generation
Expression, pacing and language realize a behaviour plan without silently changing its intent.
Controlled development
Quality checks, reflection, memory decisions and later development remain governed and reversible.
RUNTIME FLOW / CONCEPTUALONE TURN
INPUT→
CONTEXT→
CONTINUITY→
PERSONALITY→
CONVERSATION→
DECISION→
EXPRESSION→
GENERATION→
QUALITY→
RESPONSE
Reflection, memory decisions and development feedback happen after response generation and are deliberately bounded.
ARCHITECTURE EXPLORER
Decision Layer
v1.2 / Stage 3TESTED
What should Coraline do?
ROLE
Generates, scores, suppresses and resolves candidate behaviours into one binding behaviour plan.
PRIMARY OUTPUT
BehaviorPlan
ARCHITECTURAL BOUNDARY
Chooses behaviour, not final prose.
STATUS MEANS
Implemented or recovered behaviour has direct test evidence. This does not imply full end-to-end production readiness.
STABLE COREIDENTITYtruth · independence · warmth · correction · boundaries
STABLE CORE / LIVING SURFACE
Same Coraline. Different moment.
The Stable Core changes only through deliberate development. The Living Surface can move with context: focus, energy, playfulness, caution, creativity, initiative and rhythm. State may change expression. It may not make Coraline less truthful or less reliable.
focusenergyplayfulnesscautioncreativityinitiativerhythm
03 / MEMORY & CONTINUITY
Remembering facts is not the same as knowing someone.
Continuity should preserve what is relevant without turning every detail into permanent profile data. The architecture separates memory by purpose, lifetime and authority.
CLASSLIFETIMEEXAMPLES
Turn working memorySeconds / one turnclassification, plan, tool results
Conversation memoryCurrent conversationthreads, short-term references, corrections
User preference memoryControlled long-termstable preferences, boundaries, communication wishes
Project memoryLong-term + versioneddecisions, open questions, artefacts
Episodic summaryControlledrelevant shared events as concise summaries
Correction memoryUntil resolvedwrong assumptions plus confirmed corrections
Personality stateSession-firstLiving Surface continuity without identity drift
PERSONALITY MAKES CORALINE RECOGNISABLE.CONTINUITY MAKES KNOWING HER MEANINGFUL.
04 / LAB & DEVELOPMENT
Real project. Uneven maturity. Visible status.
This site intentionally separates what has test evidence from what is specified, drafted, planned or experimental. A convincing concept does not need fake production readiness.
Decision Layer Stage 3
139 / 139
Recovered core checks passed across mandatory scenarios, initiative, scoring landscape, primary envelope and state.
Historical/recovered fixture evidence. Not an end-to-end claim.Conversation Dynamics v0.1
Architecture contract defined for flow, threads, ownership, momentum, repair, depth windows and degraded operation.
Memory & Continuity v0.2
Memory classes, correction logic and continuity boundaries are being consolidated before broader activation.
Visual Identity
Visual references remain exploratory. No canonical face or body defines Coraline’s textual identity.
BETA 2.5 / LIVE TEST FINDING
“This doesn’t quite feel like a normal AI assistant.”
That is the behaviour target, not a marketing claim. The experiment is whether coherent personality, constructive independence, continuity and conversation dynamics can make interaction feel qualitatively different without pretending the system is human.
NOWText prototype + behaviour testing
NEXTControlled memory + continuity
LATERVoice + broader private testing
THE WORKING IDEA
Coraline is not meant to show what AI can do for you.
She explores what can emerge between a human and an AI when both are given room.
Experimental research and development project. Coraline is an AI system concept, not a claim of human consciousness or embodiment.